RFR (S): 8214086: [TESTBUG] Fix subgraph test cases in ReplaceCriticalClasses.java

[Ioi Lam]

Hi Jiangli,

For this test

final String subgraphInit = "initialize_from_archived_subgraph " + 
subgraphKlass;
...
if (checkSubgraph) {
      out.shouldNotContain(subgraphInit);
}

I think you should also test for the reverse condition -- if 
java/lang/Long has not been redefined, you should see the subgraphInit 
in the output.

The reason for this is shouldNotContain() is very lenient. If you made a 
mistake in formatting subgraphInit, shouldNotContain will not complain, 
and you will end up having a bad test that doesn't test what you're 
looking for.

Thanks

- Ioi



On 11/21/18 2:17 PM, Jiangli Zhou wrote:
> Please review the following test fix in ReplaceCriticalClasses.java. I 
> re-implemented the subgraph check to verify a subgraph is not used at 
> runtime if one of it's klass (who's object is contained within the 
> archived subgraph) is changed by the JVMTI agent.
>
> I added a test case for the newly archived Long cache subgraph. Since 
> both ResolvedModule and Long classes are loaded during the primordial 
> phase, '-early -notshared' are set in those test cases.
>
> As part of the change, I fixed 
> HeapShared::initialize_from_archived_subgraph to only do the logging 
> output after sbugraph(s) is successfully installed in the given Klass.
>
> webrev: http://cr.openjdk.java.net/~jiangli/8214086/webrev.00
> bug: https://bugs.openjdk.java.net/browse/JDK-8214086
>
> Tested ReplaceCriticalClasses.java locally on linux-x64. Will run 
> tier1-tier2.
>
> Thanks,
> Jiangli